In this uustat-seminar I will present a number of practical spatial prediction problems that I have been involved in over the past couple of years. The usual approach is that we split spatial variability of the measured quantity into a structural and a random component. We use some form of regression modelling for the structural variability (trend), and try to improve on the trend estimates by predicting the random component using spatial autocorrelation in the residual. Problems arise when the data are non-Gaussian, e.g. binary (presence/absense) data or count data with many zeros. Applications include: (i) spatial interpolation of sea bird densities on the Dutch part of the North Sea, using airborn strip transect bird count data and information regarding water depth and distance to the coast, (ii) spatial prediction of plant species presence in the Netherlands using the national plant databases and map information of soil characteristics, (iii) predicting aboveground biomass from field data and hyperspectral image information (an application to the La Peyne catchment, France).